Artif Intell Rev DOI 10.1007/s10462-012-9359-6

A study of the dynamic features of recommender systems Chhavi Rana · Sanjay Kumar Jain

© Springer Science+Business Media Dordrecht 2012

Abstract The extensive usage of internet is fundamentally changing the way we live and communicate. Consequently, the requirements of users while browsing internet are changing drastically. Recommender Systems (RSs) provide a technology that helps users in finding relevant contents on internet. Revolutionary innovations in the field of internet and their consequent effects on users have activated the research in the area of recommender systems. This paper presents issues related to the changing needs of user requirements as well as changes in the systems’ contents. The RSs involving said issues are termed as Dynamic Recommender Systems (DRSs). The paper first defines the concept of DRS and explores the various parameters that contribute in developing a DRS. The paper also discusses the scope of contributions in this field and concludes citing in possible extensions that can improve the dynamic qualities of recommendation systems in future. Keywords Recommender systems · Collaborative filtering · Information overload · Dynamic · Temporal

1 Introduction Recommender system research is an activated area of research that helps users in overcoming the information overload problem on the internet The first Recommender system (RS) came into existence in Resnick and Varian (1997) and since then they are evolving continuously to achieve higher degree of accuracy as well as user satisfaction. Presently several

C. Rana (B) Department of Computer Science Engineering, University Institute of Engineering and Technology, MD University, Rohtak, 124001 Haryana, India e-mail: [email protected] S. K. Jain Department of Computer Engineering, National Institute of Technology, Kurukshetra, 136119 Haryana, India e-mail: [email protected]

123

C. Rana, S. K. Jain

RS are working successfully in their domain like Amazon (Linden et al. 2003) and Netflix (Bennett and Lanning 2007), while others like Letiza (Lieberman 1995), Fab (Balabanovic and Shoham 1997) faded out with passage of time. There are several factors that are involved in the dynamics of RS success. RS are directly involved in assisting users to make decision and satisfying their current information need. In most of the cases, users are not aware of their specific need; however, they have general idea. RS can help users in making decisions about what they require at any given time if such system could provide accurate recommendations (Shani and Gunawardana 2009). This can be difficult at times when little user data is available for inferring the user evolving needs as well as system keeps on evolving with time (Chu and Park 2009). Thus, there is a need for the development of RS that could handle temporal dynamics of the user needs as well as system content and accordingly present modified recommendations to the users in real-time. Although, a decade has passed and yet the researchers have not focuses on different aspect of RS other than that of accuracy. Accuracy could not be the sole parameter for user satisfaction. Sometimes user realizes that the product is not suitable for his requirement after buying a product. Thus, a long term relationship could not be built using user purchase patterns only. Together with this, another major problem is that user preferences depends upon a range of factor like context, age, time, location, trust, and new experiences. Thus a static user profile cannot judge the preference of a user over a period of time which is commonly prevalent methodology in current recommender system (Gauch et al. 2007). A lot of research is being carried out in relation to each factor in isolation, but a comprehensive work that includes a combination of these factors and that take into account the dynamics of RS is not yet proposed. Although, researchers have worked on temporal aspect of knowledge in recommender system, still a multidimensional ubiquitous work is required (Vellino and Zeber 2007). Koren also highlighted this issue recently (Koren 2009), suggesting accuracy cannot be further enhanced in recommender system without taking into account the temporal effects. The problem is usually tackled using the existing two dimensional frameworks in recommender systems which include users and items (Adomavicius and Tuzhilin 2005). As such, the temporal evolution is presented in terms of user preferences that evolve with time or the item contents that gets changed due to addition of new items or deletion of older items (Chu and Park 2009). However, the problem of evolution is much more complex with multiple factor playing their parts with passage of time. Therefore, there is a need to extend the existing two dimensional frameworks into multidimensional factor analysis model (Adomavicius and Tuzhilin 2001) for learning the dynamics in recommender system. In this paper, we describe the concept of dynamism in RS by presenting a new category of RS called DRS. In Sect. 2 Dynamic recommender system (DRS) are defined and explained. Section 3 presents the various parameters for classification of DRS and accordingly reviews the work related to each parameter. Section 3.1 describes the scope of research in this field and some prominent aspect of an ideal recommender system. Section 3.2 concludes the paper giving future direction of research.

2 Dynamic recommender systems Recommender systems are defined as the system in which “people provide recommendations as inputs, which the system then aggregates and directs to appropriate recipients” (Resnick and Varian 1997). A number of other definitions are also given by various other researchers, each of which emphasize on RS as a tool for assisting users in finding relevant information. In this paper, a new category of RS is sketched that are dynamic in nature. The property of

123

A study of the dynamic features of recommender systems

dynamism itself is a combination of various parameters where each parameter can be the core ingredient in developing a standalone system. DRS are the systems that are able to register changes occurring in the user sphere, the system sphere as well as other environmental changes implicitly or explicitly and accordingly modify their recommendation to the users. Recommender system can be seen analogous to an advisory circle of friends that helps each other making choices in life. Taking into account user dynamics into RS may require incorporating human psychology that could play a very important part in developing such recommender system that are close to real life advisors. The human mind is extremely complex and difficult to interpret; still there are various personality attributes that could help in revealing human behavior that could then be implemented in recommender system (Hu and Pearl 2009). A novel methodology of making recommendation process more transparent as well as similar to real life recommendations should invariably involve taking into account psychological factors like trust and social network of the user (Teppan 2008). This could also be one of the key areas of working for the next generation recommender system. In addition, system side also registers a number of changes in terms of its content and goes through phases of updation and evolution. Such changes also affect the type of recommendation provided to the user and can be implemented in the RS (Lathia et al. 2009). However, the changes in the recommender systems behavior go beyond temporal factors and involve context, novelty, serendipity, real-time dynamics as well as diversity. Each one of these parameters contributes to the dynamic behavior of DRSs. As such Recommender Systems that are involved in dealing with any one of the above parameter are stated as DRSs.

3 Classification of dynamic recommender systems Dynamic Recommender Systems is not being yet established as a separate field of Recommender Systems research. Therefore, the contributing work in this field is spread along a number of spheres where different issues are being taken up in different categorization. Generally Recommender Systems are broadly divided into three categories namely content based filtering (CBF), collaborative filtering and hybrid system (Balabanovic and Shoham 1997). This categorization typically determines the core technique that is used for predicting preference and is based on the rating estimation of different users for different items. CBF builds a profile for a user based on the content features of the items previously rated by the user. The main drawback of this approach is that the recommended items are similar to the items earlier seen by the user. Mladenic (1999) presents a comprehensive survey of the commonly used text-learning techniques in the perspective of content filtering. Collaborative filtering (CF) is one of the most successful and extensively used recommender systems technologies (Schafer et al. 2001). CF examines users’ ratings to identify similarity between users on the basis of their past ratings, and then generates new recommendations based on like-minded users’ preferences (Chu and Park 2009). In the last decade, a number of approaches are developed that involves a combination of content based filtering and collaborative filtering. Such systems are termed as Hybrid recommender system. Burke conducts a detailed survey of Hybrid recommender system in which he compared different strategies underlying the development of such systems (Burke 2007). Thus, Hybrid recommender system, depicts the trend where a combination of different technique each having its own pros and cons are combined together to get the maximum benefit. It is observed that CBF when used in conjunction with standard approaches to Recommender systems such as collaborative filtering can help address some of the shortcomings of these techniques, including reliance on subjective user ratings, lack of scalability, and poor performance in the face high dimensional and sparse data (Schafer et al.

123

C. Rana, S. K. Jain

2001). However, without context knowledge and timely updation of the resulting model, such systems cannot recommend different types of complex objects that changes with time based on their underlying properties and attributes (Dai and Mobasher 2003). Recommendation techniques have a number of possible classifications (Resnick and Varian 1997; Schafer et al. 1999). All these classification are based on the kind of algorithm and the type of data used to determine user interest. Traditionally it is believed that user interests are relatively static over a period of time. Although this assumption is valid to some extent, however with the rise of internet and the kind of fast paced world we are living, things are changing at a very fast rate. Recommenders are usually built inherently dynamic to some degree, yet the rate at which user preferences (from the user point-of-view) and item list (from the system’s point of view) are changing in current scenario is beyond their coverage. This statement is being validated by a number of researchers who have worked in this area (Koren 2009; Lathia et al. 2009) Without taking into account rate, traditional recommendation is somewhat dynamic as the system is updated periodically to retain relevancy. Temporal Dimension and inclusion of other dynamic factors becomes important when tradition retraining methods cannot keep pace with the rate of change with respect to user preferences. In this paper a whole new categorization of Recommender Systems is given that are dynamic in nature and tackles this issue. This new scheme of classification is presented to explain the different facets of DRSs. The parameter over which the classification is made caters to the dynamic characteristics of Recommender Systems. These parameters cover various dimensions of the Recommender Systems that are dealt with in the research arena and are interlinked with the evolution of system as a whole over the web (Manikrao and Prabhakar 2005). The identified parameters are namely temporal context, novelty, serendipity, diversity, Dynamic environment and temporal characteristic. Each of these parameters also depict a stage in the evolution process of a Recommender System where user needs are different from other stages and thus poses a challenge for the researchers of Recommender Systems.. The computational complexities are bound to increase if one takes into account all these parameters, but there are certain evolutionary algorithms as well as other techniques that can optimize the solutions (Cho et al. 2005; Ujjin and Bentley 2003). The dynamic nature of web where items as well as users are increasing at breathtaking rates poses the foremost challenge for recommender systems research community. We will present here what is being yet proposed in this sphere related to each parameter and thus the state of the art of these 6 categories of the DRSs will be described. 3.1 Temporal changes Koren highlights the effect of temporal dynamics in RS (Koren 2009) in his research. He emphasizes on the need of including temporal changes in RS to improve the accuracy of recommendations. He also proposes a matrix factorization model that traces the time changing behavior throughout the life span of data and thus exploiting all relevant components which is in contrast with the earlier concept drift explorations where only single concept is traced. On the other hand, Lathia et al. (2009) provide a different perspective, which is a system oriented approach different from Koren (2009) user preference model. He studies the effect of retraining CF algorithm every week as a time dependent prediction problem and proposes an adaptive temporal CF technique. This technique temporally adapts the size of KNN neighborhood based on the performance measured up to current time. Also, Jambor et al. (2012) recently proposed a model of developing robust recommender system that provide up to-date recommendation over time using concepts of modern control theory. Earlier, Cho et al. (2005) on the other hand have used purchase sequence of data to improve the quality of recommendation. This purchase sequence also consists of temporal

123

A study of the dynamic features of recommender systems

parameters. An algorithm called Eigenstate is developed by Nathanson et al. (2007) that considers changes in user’s preferences with time when selecting the next item to recommend, reordering them based on the most recent rating. Another approach that incorporated temporal information to achieve better recommendation accuracy is proposed by Lee and Park (2008). They combined the two dimensions involving temporal dynamics, one proposed by Tang et al. (2003) involving product launch time and other by Ding et al. (2006) that is based on rating time, together with implicit feedback data to construct a pseudo rating data. This rating data gives more accurate recommendations. An alternative approach is given by Chu and Park (2009) where a feature-based machine learning approach to personalize recommendation of new items is proposed. This approach maintains profiles of content of interest and updates their temporal characteristics, e.g. popularity and freshness, in real-time manner. Min and Han (2005) also pointed the fact that little attention is paid to the use of time-related data in the recommendation process. They suggested a methodology that works at different stages of recommendation process for detecting user’s time-variant pattern in order to improve its performance. Also, Huang and Huang (2009) propose a sequential pattern based recommender system that predicts the customer’s time-variant purchase behavior. They studied the time decaying effect within such systems and empirically showed that this could be utilized to improve the performance. Chen and Han (2007) showed the time decaying effect of sequential pattern on the user preference within content based RS. In addition, De Pessemier et al. (2010) presents an empirical evidence that older consumption data has a negative influence on the recommendation accuracy in case of consumer centric RS. 3.2 Real-time dynamics Real-time dynamics also play a very important role in today’s fast paced life where choices of users depend on many factors susceptible to change anytime. Thus, taking into account this dynamics during online processing is necessary which is in contrast with earlier RS that works on static offline setting. This category of work is being proposed by a number of researchers. Baraglia et al. (2004) implemented a RS that collapse the two phases (online and offline) into a single online phase. This component dynamically creates links to pages that are not yet visited by a user and might be of his potential interest whereas previously the prediction about user interest are being calculated in offline phase in a 2-phase recommender system. In addition Manikrao and Prabhakar (2005) presents an “add on” for the existing RS, where it uses dynamic logic programming as an extension of answer set programming as a means for user to specify and update their models and preferences. The purposed mechanism enhances the scalability as well as accuracy of RS. Chen and Han (2007) observes that creating a private dynamic user profile (DUP) at the client side can cater to both the privacy and accuracy of the RS. He implemented a method called CRESDUP that collects, mines, discovers stores and updates private DUP at the client side. Lathia et al. (2008) propose a novel spatio-temporal model for collaborative filtering applications to suit the dynamic evolution of data. Lastly Castro-Herrera et al. (2009) implemented and evaluated a forum RS designed to handle the challenge of dynamically evolving internet forums which are characterized by a constant influx of new user and new posts. According to Khoshneshin and Street (2010) current RS are computationally expensive and therefore they work best in static offline settings. They propose an evolutionary co-clustering method that includes the new data in CF model in an online real time manner which improve predictive performance. Chandramouli et al. (2011) emphasis the need of real time recommender system in the era of social networking where user is continuously updating his profile by adding new items (e.g., news posts, Facebook postings). Therefore, they proposed StreamRec, a recommender system

123

C. Rana, S. K. Jain

architecture that leverages a stream processing system that model a recommendation system as a complex event processing (CEP) application. Kwon and Hong (2012) recently proposed a content based recommender system that utilizes real-time location-tagged information from mobile social network. 3.3 Context The importance of context in providing recommendation is being stated by a number of researchers. Various contextual features like time of the day, summer or winter season, whether user looking for self or someone else influence the decision of user to a great extent. Gonzalez et al. (2007) provides a novel perspective to improve the performance of RS through embedding emotional context with the help of SPA platform which is implemented in an intelligent learning guide. This emotional information is acquired in an incremental way to enrich recommendations in everyday life. A different approach of tackling context through the use of social network as well as user’s physical location is described by Woerndl and Groh (2007). On the other hand, Panniello et al. (2009) explores a completely new perspective by comparing the effect of pre verses post filtering approaches in context aware RS. On the basis of this study they explain that choosing the right filtering method can improve the performance of RS. Baltrunas and Ricci (2009) analyses a novel pre-filtering technique for context aware CF called item splitting. In this approach, the ratings of certain item are split based on contextual conditions and accordingly they are used in different context. They observed that the performance depends on item selection method and on the extent of influence of the contextual variables on the item ratings. The major problem in developing context aware RS is the identification of such contextual features. Said (2010) deals with the problem of context identification by framing a conceptual architecture for a context aware RS for movies. Moreover, Jancsary et al. (2010) postulates that a user’s preference for particular item depends not only on the topic and on propositional content, but also on the user’s current context. Based on his postulate, he conducts a systematic evaluation of the merit of contextual and non-propositional feature based on real life click stream and posting data. In an alternative approach, Baltrunas et al. (2012) also proposed a context based recommender system recently where they simulated the contextual situation to gather the influence of these features on user ratings. Based to the gathered data and the importance given to the relevant contextual features they build their system. Kahng et al. (2011) proposed a novel method which incorporates several contextual features into the ranking model and give more weight age to each according to their ranks. 3.4 Diversity Diversity involves presenting different types of recommendation to user which are similar in taste. Though, the feature of diversity is contrasting to accuracy, many researchers have tried to bring in congruence between the two. Ziegler et al. (2005) presents topic diversification, a novel methodology designed to balance and diversify personalized recommendation lists in order to reflect user’s complete spectrum of interests. This procedure is somewhat detrimental to accuracy and is being worked upon. Furthermore, Kwon (2008) proposes a new approach that can improve the diversity of the Top N item selection by taking into account rating variance, which can work in conjunction with any existing recommendation technique. Moreover, in this approach user can control the balance between the accuracy and diversity of recommendation through an adjusted ranking and filtering combined approaches which adjust conditions in selecting N items according to the user. Similarly, Adomavicius and

123

A study of the dynamic features of recommender systems

Kwon (2008) explore the advantage of variance in neighborhood rating in the recommendation process to overcome the accuracy/diversity tradeoff. Another researcher (Zhang and Hurley 2008; Zhang 2009) develops a model to maximize the diversity of the retrieved list while maintaining adequate similarity to the user query as a binary optimization problem. A new evaluation metric called item novelty is also proposed to compare the results. Lastly, Lathia et al. (2010) also shows that temporal diversity is an important facet of RS through a user study. They also examined the diversity of three CF over time by defining diversity metric. Moreover, they provide several methods that could be used to improve the diversity of recommendations. 3.5 Novelty Another new dimension which is being investigated lately is the idea of novelty. After a time similar items which are popular with everybody are been recommended repeatedly. This becomes very frustrating for the user at times when he looking for something new. Celma and Herrera (2008) present two methods named, item and user centric to evaluate the quality of novel recommendation. They observe that though CF recommend less novel item than CBF, user’s perceived quality is higher. This is because CF is biased towards popularity, effecting novelty and network topology while CBF is not affected at all. Park and Tuzhilin (2008) deals with the concept of novelty in a whole new way. They attempt to study the long tail problem of recommender systems where many items in the long Tail have only few ratings, thus making it hard to use them in recommender systems. They are rarely recommended but have got potential to interest user at times, finding which is not a trivial task. Abbassi et al. (2009) examines the case of over-specialization in recommender systems, which results from returning items that are too similar to those previously rated by the user. They develops an algorithm outside the box (OTB), that attempts to identify regions that are underexposed to users, by taking some risk to help users make fresh discoveries, while maintaining high relevance. On the other hand, Vargas and Castells (2011) noted that there is lack of well defined evaluation metrics in this area that take into account their ranking. Therefore, they proposed a framework built upon three ground concept namely choice, discovery and relevance and generalizes several state of the art metrics using them. Vargas (2011) also presented the application of intent oriented Information Retrieval diversity techniques to the RS field, which is still in progress together with the formalization of novelty and diversity metrics for their evaluation. 3.6 Serendipity Lastly comes the concept of serendipity. Serendipity is a tendency for making fortunate discoveries while looking for something unrelated (http://dictionary.cambridge.org/dictionary/ british/serendipity). As explained by Herlocker et al. (2004) there is a surprise element attached to it that differs it from the novelty feature. Due to the explosive growth of web and henceforth the choices emerging from it, users are looking for adventurous encounters, in addition to the normal requirement. Although the effect of serendipity in RS is being studied by very few researchers, it is gaining popularity lately. One of prominent work in this direction is carried by Iaquinta et al. (2010). He stated that there are some context in which user requires unsearched but still useful items or pieces of information. He proposes a hybrid RS that joins a CBF and serendipity heuristics in order to mitigate the overspecialization problem with surprise suggestion. In addition, Ge et al. (2010) emphasis on the need to evaluate the quality of RS beyond accuracy. They analyze the role of coverage and serendipity as

123

C. Rana, S. K. Jain Table 1 Classification of dynamic recommender systems Sr. no.

Parameter

Description

References

1.

Temporal changes

This parameter takes into account the effect of time aspect on the RS. They recommend when rather than what

2.

Online processing

This category describes what is widely termed as the real-time behavior of RS

3.

Context

It is the parameter which describes a particular state of the user or the environment at any given period of time

4.

Novelty

5.

Serendipity

6.

Diversity

Quality of being striking, new and original that discover new items for the user It is propensity for making fortunate discoveries while looking for something unrelated. Differs from novelty in the sense that a surprise element is attached It defines the variety in choices breaking the barrier of similarity

Koren (2009), Lathia et al. (2010), Jambor et al. (2012), Cho et al. (2005), Nathanson et al. (2007), Lee and Park (2008), Ding et al. (2006), Tang et al. (2003), Chu and Park (2009), Min and Han (2005), Huang and Huang (2009), Chen and Han (2007), De Pessemier et al. (2010) Baraglia et al. (2004), Manikrao and Prabhakar (2005), Chen and Han (2007), Lathia et al. (2008), Castro-Herrera et al. (2009), Khoshneshin and Street (2010), Kahng et al. (2011), Chandramouli et al. (2011), Kwon and Hong (2012) Gonzalez et al. (2007), Woerndl and Groh (2007), Panniello et al. (2009), Baltrunas and Ricci (2009), Said (2010), Jancsary et al. (2010), Kahng et al. (2011) Celma and Herrera (2008), Park and Tuzhilin (2008), Abbassi et al. (2009), Vargas and Castells (2011) Iaquinta et al. (2010), Ge et al. (2010), Oku and Hattori (2011)

Ziegler et al. (2005), Kwon (2008), Adomavicius and Kwon (2008), Zhang and Hurley (2008), Zhang (2009), Lathia et al. (2010)

indicators of recommendation quality, and presents novel ways to measure them as well. Lin and Chen (2011) focused on utilizing the interactive information present in the social network of each user for providing serendipitous recommendations. Oku and Hattori (2011) propose a Fusion-based Recommender System which finds new serendipitous items that have mixed features of two user-input items, produced by mixing the two items together. The Table 1 illustrates in brief the work mentioned above according to each parameter.

4 Scope of contribution User satisfaction is the ultimate goal of any system that is developed for the prime purpose of user interaction. Financial benefits are attached to this goal and an increased user satisfaction will ultimately convert into long term relationship and thus increased revenues (Guy et al. 2010). To achieve this goal there are a number of factors that come into play, foremost of which is the user interface that enriches the experience of incoming user on the website

123

A study of the dynamic features of recommender systems

(Said 2010; Shneiderman 1997). Earlier systems provided a simple user interface that asks too much explicitly from the user which at times becomes very annoying. Thus, lately there is too much emphasis on automatic Recommender Systems that can derive its inferences from implicit user data like browsing behavior and click streams. Currently most of the recommender systems are automated (Pu et al. 2008). Although, automation is helpful for the user to some extent, it is not the solution for determining user changing requirement. Therefore, together with automation, some amount of user control is also needed so that users can filter out useless recommendation as well as define his context and motives. A sense of balance is needed to achieve the desired level of user satisfaction by implementing a system which is neither too intrusive nor too rigid (Pronk et al. 2007). The second factor which effects user satisfaction in a predominant way is the accuracy of recommendation provided. The accuracy of recommendations depicts the needs of each particular user and provides a solution which fulfills those needs. The accuracy of recommendations can be enhanced through efficient algorithm as well as enriched data. Few researchers have also emphasized on an accurate taxonomy in term of document topic labels (Ding et al. 2006) which impart a big positive effect on the results of a particular algorithm. Thirdly, evaluation of Recommender systems is mostly carried out on a set number of accuracy metrics which are derived from information retrieval arena and machine learning techniques (Ge et al. 2010). These metrics cannot depict the performance of RS in a highly dynamic environment of web. Thus, there is a huge need to develop evaluation metrics for judging the dynamics of recommendation process together with the performance (Burke 2010). Lastly, a number of assisting features like ease of integration and administration as well as updating recommendations in real time represent other major requirements of ideal recommender systems. It is observed that while searching is about finding answers to what you already want to know, recommender systems opens up a set of possibilities of what you might want to know (Guy et al. 2010). Thus, novelty, serendipity and diversity are going to become very important aspect of Recommender Systems in future that will in turn define how the evolution process of user interest could be satisfied with the help of them. Therefore, a number of possibilities and challenges mark the future of DRSs research which opens up a very interesting era of human computer interaction in a whole new way.

5 Future directions Dynamic Recommender Systems is a fairly new concept. The term is not being yet used in the widely used taxonomy of Recommender Systems. Though the need to include temporal evolution of data as well as user preference in the Recommender Systems has been repeatedly stated (Chu and Park 2009; Gauch et al. 2007), few researchers have worked in this area. This dynamism is parameterized in various forms namely temporal context, novelty, serendipity, diversity, dynamic environment and temporal characteristic. An integrated view of this field has not been attempted before. The challenges faced by recommender systems in the highly dynamic environment of the web though have been dealt with, in an individual manner; a holistic approach is not yet proposed. There is a huge need to find ways in which user context and expectation could be judged by combining explicit information which is collected by giving user the flexibility to choose and guide the recommendation process as well as by automatically discovering new knowledge through the use of implicit data in the recommendation process (Fernando et al. 2010). Motivating user to get involved in the recommendation process needs to be dealt with first, to bring this change. A whole new generation of recommender systems that are

123

C. Rana, S. K. Jain

using social tagging for improved user participation are also being researched upon widely (Marinho et al. 2011). Some researchers also suggest studying various other dimensions of recommendation process beyond accuracy and rating that could enhance novelty and diversity (Ge et al. 2010). This can make a huge impact on improving the dynamics of RS. A sense of balance is required while giving recommendation, that combines the search capabilities as well as gives prompt recommendation and as such provide user with a whole new interface of experiencing web data. For example, Hurley and Zhang (2011) presented a review on the need of diversity research in recommender system and suggested that the trade-off between diversity and accuracy could be taken as binary optimization problem, with an input control parameter allowing explicit tuning of this trade-off. There is need to do a comprehensive research on the core module that could implement the dynamics of RS, which will continue to improve upon in various dimensions and not just accuracy. A number of evolutionary algorithm as well as other modifications of standard collaborative filtering algorithm are being continuously devised to deal with the dynamism of the whole process so far (Geyer-Schulz 2000; Demir et al. 2007). The advancement in the social networks over the internet also provided a new direction to these efforts (Carmagnola et al. 2009). Zhou et al. (2011) conducted a review of the state of the art of the existing technologies for building personalized recommender systems in social networking environment. Explanation in the recommender systems interfaces as well as human psychology factors are also expected to contribute a lot in the development of new solution to bring up DRSs a new face in the coming decade (Tintarev and Masthoff 2007). Konstan and Riedl (2012) recently presented a survey of major breakthrough in recommender system from user perspective and algorithmic research stating future challenges and relevance to real world applications. World Wide Web continues to evolve just like the human race and this evolution is going to bring up new challenges as well as opportunities for human race to benefit from it. 6 Conclusion The field of Recommender systems research is matured with lots of algorithms being developed, tested and compared with each other. However, it maintains its charm with new discoveries and some completely new dimensions of work. The contribution of this paper is in bringing out together; the work from a number of spheres that share a common referral point which is dynamism. The paper classifies the current RS research into a number of spheres based on parameters namely temporal context, novelty, serendipity, diversity, dynamic environment and temporal characteristic. The work related to each sphere is also being discussed. Our paper focuses on the need for innovative solutions that provides enhanced user satisfaction as well as system performance. These solutions go beyond the current parameters of accuracy and efficiency and bring a novel experience to the user while browsing World Wide Web. Additionally, a number of future research directions are also suggested that emphasize on getting past the human inertia, breaking barriers of understanding the context of user as well as making real-time conversation with user to infer its needs at the right time in the right manner.

References Abbassi Z, Amer-Yahia S, Lakshmanan LVS, Vassilvitskii S, Yu C (2009) Getting recommender systems to think outside the box. In: Proceedings of the third ACM conference on recommender systems RecSys 09, ACM Press, pp 285–288

123

A study of the dynamic features of recommender systems Adomavicius G, Tuzhilin A (2005) Toward the next generation of recommender systems: a survey of the state-of-the-art and possible extensions. IEEE Trans Knowl Data Eng 17(6):734–749 Adomavicius G, Tuzhilin A (2001) Extending recommender systems: a multidimensional approach. In: Proceedings of the international joint conference on artificial intelligence IJCAI01 workshop on intelligent techniques for web personalization (ITWP2001). Seattle, pp 4–6 Adomavicius G, Kwon Y (2008) Overcoming accuracy-diversity tradeoff in recommender systems: a variance-based approach. In: Proceedings of the 18th workshop on information technology and systems (WITS’08), Paris Balabanovic M, Shoham Y (1997) Fab: content-based, collaborative recommendation. Commun ACM 40(3): 66–72 Baraglia R, Merlo F, Silvestri F (2004) An effective recommender system for highly dynamic and large web sites. In: Proceedings of the 8th European conference on principles and practice of knowledge discovery in databases (PKDD ’04). Springer, New York, pp 531–533 Baltrunas L, Ricci F (2009) Context-based splitting of item ratings in collaborative filtering. In: Proceedings of the third ACM conference on recommender systems (RecSys 09), New York, pp 245–248 Baltrunas L, Ludwig B, Peer S, Ricci F (2012) Context relevance assessment and exploitation in mobile recommender systems. Pers Ubiquitous Comput 16(5):507–526 Bennett J, Lanning S (2007) The netflix prize. In: Proceedings of KDD cup and workshop Burke R (2007) Hybrid web recommender systems. In: Brusilovsky A, Kobsa A, Nejdl W (eds) The adaptive web, vol 4321. Springer, Berlin, pp 377–408 Burke R (2010) Evaluating the dynamic properties of recommendation algorithms. In: 4th ACM conference on recommender systems, pp 225–228 Carmagnola F, Vernero F, Grillo P (2009) SoNARS: a social networks-based algorithm for social recommender systems. In: Geert- Jan Houben, Gord Mccalla, Fabio Pianesi, and Massimo Zancanaro (eds) Proceedings of the 17th international conference on user modeling, adaptation, and personalization: formerly UM and AH (UMAP ‘09). Springer-Verlag, Berlin, Heidelberg, pp 223–234 Castro-Herrera C, Cleland-Huang J, Mobasher B (2009) A recommender system for dynamically evolving online forums. In: Proceedings of the third ACM conference on recommender systems (RecSys 09), New York, pp 213–216 Celma Ò, Herrera P (2008) A new approach to evaluating novel recommendations. Proceedings of the 2008 ACM conference on recommender systems (RecSys 08), Lausanne, pp 179–186 Chandramouli B, Levandoski JJ, Eldawy A, Mokbel MF (2011) StreamRec: a real-time recommender system. In: Proceedings of the 2011 ACM SIGMOD international conference on management of data (SIGMOD ’11). ACM, New York, pp 1243–1246 Chen T, Han W (2007) Content recommendation system based on private dynamic user profile. Mach Learn 4:2112–2118 Chiu Y-S, Lin K-H, Chen J-S (2011) A social network-based serendipity recommender system. In: International symposium on intelligent signal processing and communications systems ISPACS, pp 1–5 Cho YB, Cho YH, Kim SH (2005) Mining changes in customer buying behavior for collaborative recommendations. Expert Syst Appl 28(2):359–369 Chu W, Park ST (2009) Personalized recommendation on dynamic content using predictive Bilinear Models. In: 18th international WWW conference, Madrid, pp 691–706 Dai H, Mobasher B (2003) A road map to more effective web personalization: integrating domain knowledge with web usage mining. In: Proceedings of the international conference on internet computing, Las Vegas, pp 1–8 De Pessemier T, Dooms S, Deryckere T, Martens L (2010) Time dependency of data quality for collaborative filtering algorithms. In: Proceedings of the 2010 ACM conference on recommender systems (RecSys ‘10), New York, pp 281–284 Demir GN, Sima Uyar S, Oguducu S (2007) Graph-based Sequence clustering through multiobjective evolutionary algorithms for web recommender systems. In: Thierens D (ed) Computer engineering, vol 2, pp 1943–1950 Ding Y, Li X, Orlowska M (2006) Recency-based collaborative filtering. In: Gillian Dobbie and James Bailey (eds) Proceedings of the 17th Australasian Database Conference (ADC ’06), vol 49. Australian Computer Society, Inc., Darlinghurst, Australila, pp 99–107 Fernando D, Enrique C, Pedro G, Campos G, Alejandro B (2010) Movie recommendations based in explicit and implicit features extracted from the Filmtipset dataset. In: Proceedings of the workshop on context-aware movie recommendation (CAMRa ’10). New York, pp 45–52 Gauch S, Speretta M. Chandramouli A, Micarelli A (2007) User profiles for personalized information access. In: Peter Brusilovsky, Alfred Kobsa, and Wolfgang Nejdl (eds) The adaptive web. Lecture notes in computer science. Springer-Verlag, Berlin, Heidelberg, 4321:54–89

123

C. Rana, S. K. Jain Geyer-Schulz A (2000) myvu: a next generation recommender system based on observed consumer behavior and interactive evolutionary algorithms. In: Gaul W, Opitz O, Schader M (eds). Data analysis: scientific modeling and practical applications, studies in classification, data analysis, and knowledge organization. Springer, Heidelberg, pp 447–457 Gonzalez G, De La Rosa JL, Montaner M, Delfin S (2007) Embedding emotional context in recommender systems. In 2007 IEEE 23rd international conference on data engineering workshop, pp 845–852 Guy I, Jaimes A, Agull P, Moore P, Nandy P, Nastar C, Schinzel H (2010) Will recommenders kill search?: recommender systems—an industry perspective. In: Proceedings of the fourth ACM conference on recommender systems (RecSys ’10). ACM, New York, pp 7–12 Ge M, Delgado-Battenfeld C, Jannach D (2010) Beyond accuracy: evaluating recommender systems by coverage and serendipity. In: Proceedings of the fourth ACM conference on recommender system, pp 257–260 Herlocker JL, Konstan JA, Terveen LG, Riedl JT (2004) Evaluating collaborative filtering recommender systems. ACM Trans Inf Syst 22(1):5–53 Huang C-L, Huang W-L (2009) Handling sequential pattern decay: developing a two-stage collaborative recommender system. Electron Commer Res Appl 8(3):117–129 Hu R, Pearl P (2009) Acceptance issues of personality-based recommender systems. In: Proceedings of the third ACM conference on recommender systems (RecSys ’09), New York, pp 221–224 Hurley N, Zhang M (2011) Novelty and diversity in top-N recommendation— analysis and evaluation. ACM Trans Internet Technol 10(4):30 Iaquinta L, Gemmis M, Lops P, Semeraro G, Molino P (2010) Can a recommender system induce serendipitous encounters? E-commerce, IN-TECH, Vienna, pp 229–246 Jambor T, Wang J, Lathia N (2012) Using control theory for stable and efficient recommender systems. In: Proceedings of the 21st international conference on world wide web (WWW ’12). ACM, New York, pp 11–20 Jancsary J, Neubarth F, Trost F (2010) Towards context-aware personalization and a broad perspective on the semantics of news articles. In: Proceedings of the fourth ACM conference on Recommender systems (RecSys ’10). New York, pp 289–292 Kahng M, Lee S, Lee S (2011) Ranking in context-aware recommender systems. In: Proceedings of the 20th international conference companion on world wide web (WWW’11). ACM, New York, pp 65–66 Khoshneshin M, Street WN (2010) Incremental collaborative filtering via evolutionary. In: 4th ACM conference on recommender systems, pp 325–328 Konstan JA, Riedl J (2012) Recommender systems: from algorithms to user experience. User Model User Adapt Interact 22(1–2):101–123 Koren Y (2009) Collaborative filtering with temporal dynamics. In: Proceedings of the 15th ACM SIGKDD international conference on knowledge discovery and data mining (KDD ’09). New York, pp 447–456 Kwon Y (2008) Improving top-n recommendation techniques using rating variance. In: Proceedings of the 2008 ACM conference on recommender systems (RecSys 08), Lausanne, pp 307–310 Kwon H-J, Hong K-S (2012) Personalized real-time location-tagged contents recommender system based on mobile social networks. In: IEEE international conference on consumer electronics ICCE, pp 558–559 Lathia N, Hailes S, Capra V (2008) kNN CF: a temporal social network. In: Proceedings of the 2008 ACM conference on recommender systems, Lausanne, pp 227–234 Lathia N, Hailes S, Capra L, Amatriain X (2010) Temporal diversity in recommender systems. In: Proceedings of the 33rd international ACM SIGIR conference on Research and development in information retrieval (SIGIR ’10). New York, USA, pp 210–217 Lathia N, Hailes S, Capra L (2009) Temporal collaborative filtering with adaptive neighbourhoods. In: Proceedings of the 32nd international ACM SIGIR conference on research and development in information retrieval (SIGIR 09), pp 796–797 Lee T, Park Y (2008) A time-based approach to effective recommender systems using implicit feedback. Expert Syst Appl 34(4):3055–3062 Lieberman H (1995) Letizia: an agent that assists web browsing. In: Proceedings of the fourteenth international joint conference on artificial intelligence (IJCAI-95). Morgan Kaufmann publishers Inc., San Mateo, pp 924–929 Linden G, Smith B, York J (2003) Amazon.com recommendations: item-to-item collaborative filtering. IEEE Internet Comput 7(1):76–80 Manikrao US, Prabhakar TV (2005) Dynamic selection of web services with recommendation system. In: International conference on next generation web services practices (NWeSP05), pp 117–121 Marinho Balby L, Nanopoulos A, Schmidt-thieme L (2011) Social tagging recommender systems. In: Francesco Ricci et al.(ed) Recommender systems handbook. Springer, pp 615–644 Min SHm, Han I (2005) Detection of the customer time-variant pattern for improving recommender systems. Expert Syst Appl 28(2):189–199

123

A study of the dynamic features of recommender systems Mladenic D (1999) Text-learning and related intelligent agents: a survey. IEEE Intell Syst Appl 14(4):44–54 Nathanson T, Bitton E, Goldberg K (2007) Eigentaste 5.0: constant-time adaptability in a recommender system using item clustering. In: Proceedings of the 2007 ACM conference on recommender systems, pp 149–152 Oku K, Hattori F (2011) Fusion-based recommender system. In: ACM RecSys workshop on novelty and diversity in recommender systems (DivRS), pp 1–7 Panniello U, Tuzhilin A, Gorgoglione M, Palmisano C, Pedone A (2009) Experimental comparison of prevs. post-filtering approaches in context-aware recommender systems. In: Proceedings of the third ACM conference on recommender systems, pp 3–6 Park Y-J, Tuzhilin A (2008) The long tail of recommender systems and how to leverage it. In: Proceedings of the 2008 ACM conference on recommender systems (RecSys 08), Lausanne, pp 11–18, October 2008 Pronk V, Verhaegh W, Proidl A, Tiemann M (2007) Incorporating user control into recommender systems based on naive bayesian classification. In: Proceedings of the 2007 ACM conference on recommender systems (RecSys 07), Minneapolis, pp 73–80 Pu P, Bridge DG, Mobasher B, Ricci F (eds) Proceedings of 2008 ACM conference on recommender systems 2008 Resnick P, Varian HR (1997) Recommender systems. Commun ACM 40(3):56–58 Said A (2010) Identifying and utilizing contextual data in hybrid recommender systems. Im: Proceedings of the 2010 ACM conference on recommender systems, Barcelona, pp 365–368 Schafer JB, Konstan JA, Riedl J (2001) Ecommerce recommendation application. Data Min Knowl Discov 5(1/2):115–153 Schafer JB, Konstan J, Riedi J (1999) Recommender systems in e-commerce. In: Proceedings of the 1st ACM conference on electronic commerce (EC’99). ACM, New York, pp 158–166 Shani G, Gunawardana A (2009) Evaluating recommendation systems. In: Ricci F, Rokach L, Shapira B, Kantor PB (eds) Springer, Berlin, pp 1–41 Shneiderman B (1997) Direct manipulation for comprehensible, predictable, and controllable user interfaces. In: Proceedings of international conference on intelligent user interfaces (IUI97), Orlando, pp 33–39 Tang TY, Winoto P, Chan K (2003) Scaling down candidate sets based on the temporal feature of items for improved hybrid recommendations. In: Bamshad Mobasher and Sarabjot Singh Anand (eds) Proceedings of the 2003 international conference on intelligent techniques for web personalization (ITWP’03). Springer-Verlag, Berlin, Heidelberg, pp 169–186 Teppan EC (2008) Implications of psychological phenomenons for recommender systems. In: Proceedings of the 2008 ACM conference on Recommender systems RecSys 08, Lausanne, pp 323–326 Tintarev N, Masthoff J (2007) A survey of explanations in recommender systems. 2007 IEEE 23rd international conference on data engineering workshop, vol 7, pp 801–810 Ujjin S, Bentley PJ (2003) Particle swarm optimization recommender system. In: Proceedings of the IEEE swarm intelligence symposium 2003 (SIS 2003), Indianapolis, pp 124–131 Vargas S (2011) New approaches to diversity and novelty in recommender systems. In: Fourth BCS-IRSG symposium on future directions in information access (FDIA 2011), Koblenz, 31 August 2011 Vargas S, Castells P (2011) Rank and relevance in novelty and diversity metrics for recommender systems. In: Proceedings of the fifth ACM conference on recommender systems (RecSys ’11). ACM, New York, pp 109–116 Vellino A, Zeber D (2007) A hybrid, multi-dimensional recommender for journal articles in a scientific digital library. In: 2007 IEEEWICACM international conferences on web intelligence and intelligent agent technology workshops, pp 111–114 Woerndl W, Groh G (2007) Utilizing physical and social context to improve recommender systems. In: 2007 IEEEWICACM international conferences on web intelligence and intelligent agent technology workshops, pp 123–128 Zhang M (2009) Enhancing diversity in top-N recommendation. In: Proceedings of the third ACM conference on recommender systems (RecSys 09), pp 397–400 Zhang M, Hurley N (2008) Avoiding monotony: improving the diversity of recommendation lists. In: Proceedings of the 2008 ACM conference on recommender systems (RecSys ’08). ACM, New York, pp 123–130 Zhou X, Xu Y, Li Y, Josang A, Cox C (2011) The state-of-the-art in personalized recommender systems for social networking. Artif Intell Rev 37(2):119–132 Ziegler C-N, Mcnee SM, Konstan JA, Lausen G (2005) Improving recommendation lists through topic diversification. In: Proceedings of the 14th international conference on world wide web WWW 05, pp 22–32

123